A blog related to original AI artworks. I make AI artworks using coined words and emoji as prompts.
I asked ChatGPT what it would be like to make AI art the way I do it now. I did not get the answer I was hoping for, but when I finally asked what would happen if I used a word that was not in the training data, I got a satisfactory answer that it was difficult to predict.
At first, I asked, “Can you make AI art with a coined spell?” The summary of the answer is, “It requires images containing coined words as training data.” I think it is a common sense answer, but what I have been doing lately is making AI art with words that are not in the training data, so I haven not answered the intended question.
So, I asked, “What happens if you create AI art using words that are not included in the training data as spells?” The short answer is, “It is hard to predict what the outcome will be.” The normal way of making AI art in the world is predictable for ChatGPT. On the other hand, what I am doing is unpredictable, that is, it is new, and I think this is a satisfactory answer.
Finally, I will post the full text of ChatGPT’s answer. “It is possible to create AI art using words that are not included in the training data as spells, but it is difficult to predict what the results will be. If you use a word that is not included in the training data, you have not learned what the word means or what image it represents, so if you use it as a spell, what it is It is also hard to predict when an image will be generated. Also, AI art obtained by using words not included in the training data as spells can be difficult for the viewer to understand. Therefore, AI art obtained by using words not included in the training data as spells requires trial and error. ”
Note: This article is a revised and translated version of an article in Brog from Kanada (in Japanese).